Slashdot Mirror
Robots Modeled On Ancient Fish Help Researchers Study Origins of Extinct Species
Hallie Siegel writes: Hypotheses about the evolution of traits in ancient species are difficult to test, as the animals have often been extinct for thousands or millions of years. In this article, researchers at Vassar College describe how a population of physical, free-swimming robots modeled after ancient fish evolved vertebrae under selection pressures for predator avoidance and foraging ability, showing how evolutionary robotics can be used to help biologists test hypotheses about extinct animals.
I share the same sediment.
We tend to think that each trait evolved was there for some sort or benefit. However a trait can happen just as long as it didn't create a significant hindrance.
So if you take the normal bell curve. On the left side are traits that are harmful, on the right side you get traits that are helpful.
Lets say 2 sigma to the left means the disadvantage is large enough to cause them to die before they can reproduce. So the disadvantages that are 1 sigma to the which are still harmful will pass on to the next generation then you got traits that fall on the average which are neutral. Over time you get on the average better, however not all traits have a use initially or ever.
Lets use humans who have acne. There is no real advantage to it, it is slightly disfigurating so it could reduce your chances of mating, you can get wounds that cause infections... But that is so minor that passing the genes that cause acne over and over.
Now lets say in a million year the acne had evolved over time that causes the skin to be thicker, and act as a form of armor. But right now it is just an inconvenience, or after a million years it can go away, or just still plague people.
If something is so important that you feel the need to post it on the internet... It probably isn't that important.
You are incorrect, sir. It is a hint of ding.
Heyo! This is Sonia Roberts, lead author on the publication.
We used physical robots at Vassar and collaborated with groups at Lafayette College to create a physics engine-based computer simulation.
It's a common misconception that writing some computer code is necessarily faster and less expensive than building robots. We got our robotic system up and running some preliminary experiments a couple of years before the computer simulation was finished. Flexible, composite solids bending in fluids are not easy to model to the degree of accuracy that we were looking for.
It's also a good idea generally if you're making an argument based on biomechanics, as we were, to have some sort of physical model to check that your results did not arise from a discrepancy between your physics engine and real-world physics ...
I read the entire paper, thinking OMG, how much effort they went to building those toys! There is no way that would produce useful results, especially in just 10 generations. The model is simply too crude, and 10 generations is absurdly brief to expect useful results with this set up.
I have spent years playing with genetic algorithms, and I have learned a great deal from them. You can never underestimate the effects of chance, and I found it takes thousands of generations to continuously break out of local maxima and move forward to achieve some substantial and interesting result.
For example, for a long time I was obsessed with breading some "super being". I would run some population over tens of thousands of generations until one of my organsims achieved the highest score ever seen, and imagine I had achieved my dream of a super hero. Then later, the colony would start "de-evolving" (they always do, in the end), I would drop that organism back into the game with the same genes that scored so high, and find that its performance now was just mediocre. The problem is, that it seems half the time I am breeding for Luck, and that turns out not to be an inheritable trait :)
Let me clarify that a bit: My experiments always involved organisms that wander around my computer screen looking for bits of food. There may be predators in some experiments, but mostly not because predators add so much turbulence. The organisms are in competition with each other, and those who score best go on to bread the next generation. I always employed small populations of 20 to 60 organisms, and a great variety of schemes in the design of their world and for dropping food into their world. During any given generation, some will score highly, and some will score poorly. Turns out, quite often this is just as much due to chance as it is to skill, and evolution goes sideways. Certainly over time I have had many "successes", however you might describe that, where organisms have developed certain traits exactly as you would expect or even exceeding expectation. However, in none of my experiments did I ever expect or observe some result in just 10 generations.
I would suggest that an experiment in software could be developed that would shed more light on the evolution of vertebrate then your toys could ever show, and I suggest that it doesn't have to have a complex physics engine that takes many months to develop. Compared to how crude you little toy fish were, I would bet I could develop a useful simulation in a matter of days. It doesn't have to model the physics of water. In fact, the effectiveness of different numbers of vertebrae in your model could be measured empirically and the resulting data transferred to the program, etc... The feeding and prey avoidance can be far more simply and effectively modeled in software. In short, I just don't think building little toy fish is at all practical to answer the questions about evolution that you pose. However, it seems you have been well taught and have conducted very good work that will serve you well when you go on to research something more practical.
If you are interested here is one of my neural network simulations that evolves via a genetic algorithm. They develop social behaviour - specifically, they learn to take turns: http://www.tropicalcoder.com/N...
Experiments like this make me believe that "survival of the fittest" is an overly simplistic statement about the mechanisms of natural selection, and that you cannot isolate the individual from the group. More and more, I forget about my original dreams of creating super beings and think about evolution at the level of the species, rather than the individual.